Back to Projects
Computer Graphics Research
Retrieval-Based PBR Textures
An end-to-end pipeline combining mesh segmentation and multimodal retrieval to assign realistic PBR materials to 3D assets.
Tay Khai Li, Vincent Limardi, Kai-Wen Hsiao, Shih-Hsuan Hung, Hung-Kuo Chu
Results
Wooden Bucket
Overview
Research project at the CGV & VM Lab building an end-to-end pipeline that segments 3D meshes and uses multimodal retrieval to assign physically-based rendering (PBR) material maps (diffuse, roughness, metallic, normal) to 3D assets, producing more realistic surface appearance than existing style-transfer-based texturing approaches. Explore the interactive comparison viewer below to see the retrieved PBR materials applied to several 3D models side-by-side with the originals.
Technologies
Mesh SegmentationMultimodal RetrievalComputer Vision3D GraphicsPython
Key Outcomes
- Top 10 NTHU CS project finalist
- Nominated for an Outstanding Paper Award at CVGIP 2026 (Taiwan)